Expertise in video understanding tasks

My work is mainly devoted to understand human-to-object and human-to-human interactions from untrimmed videos, with focus on unsupervised and weakly-unsupervised approaches that reduce or eliminate the need of annotations during training, hence providing methods that are more scalable and suitable for real world applications. Example of tasks are action recognition, action segmentation, action anticipation, motion segmentation, turn-taking prediction during social interactions, social interaction recognition. I have also experience in developing video understanding methods methods tailored to concrete health and robotic applications.